EPUB → MP3 — Audiobook Converter for Windows

Turns an EPUB book into one continuous MP3 audiobook, narrated by natural-sounding Microsoft neural voices, with chapter markers, book metadata and cover art embedded in the file.

A small Tkinter desktop app with a command-line mode, packaged as a single .exe with PyInstaller. The build recipe is in Building from source.

---

Features

• One MP3 file per book — easy to copy to a phone, USB stick or car radio.
• Three audio quality presets — 48 kbps at 24 kHz (native voice format, smallest), or 64 / 128 kbps at 44.1 kHz (standard MP3 that plays in every player, including older car stereos).
• Natural neural voices — 300+ Microsoft Edge voices, 47 of them English (US, UK, AU, CA, IN, IE, NZ, ZA).
• Works offline too — falls back to the Windows voices installed on your PC if the internet drops (or pick an offline voice from the start).
• Chapter markers (ID3 CHAP/CTOC) — jump between chapters in VLC, foobar2000, Poweramp, etc.
• Title, author, year, cover read from the EPUB and written into the MP3 tags.
• Speed control from 0.5× to 2×.
• Smart text handling — removes page numbers, Project Gutenberg boilerplate, duplicate chapter headings and illustration placeholders; splits text at sentence boundaries for natural pauses.
• Resumable-free and safe — temp files are cleaned up, a failed run never deletes an existing MP3.
• Self-contained build — PyInstaller produces one executable that needs no Python on the target machine.

Quick start

Requires Python 3.13 on Windows — the offline voices go through SAPI via pywin32.

python -m venv .venv
.venv\Scripts\pip install -r requirements.txt
.venv\Scripts\python src\launcher.py

Or as a module:

set PYTHONPATH=src
.venv\Scripts\python -m epub2mp3

Then:

1. 1 · Book → Browse… and pick your .epub.
2. 2 · Voice → choose a narrator and an audio quality (the list shows online and offline voices).
3. 3 · Save as → the file name is filled in automatically; change it if you like.
4. Press Convert to MP3 and watch the progress bar. When it finishes, use Open folder.

Long books take a while: roughly 5× faster than real time with online voices, so a 3-hour audiobook needs about 35–45 minutes of conversion.

Audio quality

Pick Quality in the Voice panel:

Preset	Bitrate	Sample rate	Size per hour	Notes
Standard	48 kbps	24 kHz	≈ 21 MB	Native voice format, best quality per byte
Compatible	64 kbps	44.1 kHz	≈ 28 MB	Standard MP3, plays in every player
Maximum	128 kbps	44.1 kHz	≈ 56 MB	Highest bitrate and widest compatibility

The neural voices are synthesised at 24 kHz. The two 44.1 kHz presets are produced with a polyphase resampler and the LAME encoder, so they are ordinary standard MP3 files rather than 24 kHz streams. All presets are mono 16-bit audio with ID3v2.3 tags, cover art and chapter markers; a 10-hour book is 210 MB (standard), 280 MB (compatible) or 560 MB (maximum).

Voices

Online (neural, best quality) — requires an internet connection. The catalogue is downloaded from Microsoft and cached for a week; press Refresh to update it. Recommended English voices:

Voice	Accent	Style
en-US-AriaNeural	US	warm, natural (default)
en-US-JennyNeural	US	friendly
en-US-AvaNeural	US	expressive
en-US-AndrewNeural	US	deep, calm
en-US-GuyNeural	US	male, clear
en-GB-SoniaNeural	British	soft
en-GB-RyanNeural	British	male
en-AU-NatashaNeural	Australian	bright

Tick Show voices in all languages to narrate books in other languages.

Offline — whatever is installed in Windows (Settings → Time & language → Speech). These sound more robotic but work with no internet at all.

Command line (optional)

The same program converts without opening a window — handy for scripting:

python src\launcher.py "C:\Books\Dune.epub" --convert --out "C:\Books\Dune.mp3"
python src\launcher.py "C:\Books\Dune.epub" --convert --bitrate 128 --out "C:\Books\Dune.mp3"
python src\launcher.py "C:\Books\Dune.epub" --convert --offline --out "C:\Books\Dune.mp3"

Option	Meaning
--convert	run headless, no window
--out PATH	output MP3 (default: next to the EPUB)
--voice ID	e.g. en-US-AriaNeural
--quality NAME	standard, compatible or maximum
--bitrate N	shortcut for --quality: 48, 64 or 128
--offline	use a Windows offline voice
--rate N	speed in percent, -50 … 100
--no-fallback	fail instead of switching to an offline voice
--quiet	only errors
--version	print the version

Headless runs started without a console (for example from a shortcut) write their log to %LOCALAPPDATA%\EpubToMp3\logs\headless.log.

Where things are stored

• Settings (last voice, speed, folders): %LOCALAPPDATA%\EpubToMp3\settings.json
• Voice cache: %LOCALAPPDATA%\EpubToMp3\voices.json
• Temporary audio during a conversion: %TEMP%\epub2mp3-* (deleted automatically)

Nothing else is written to the registry.

Troubleshooting

“Windows protected your PC” — an exe you build yourself is not code-signed, so SmartScreen shows this warning the first time. Click More info → Run anyway.

Online voice errors / “could not load online voices” — check the internet connection. A proxy or firewall that blocks speech.platform.bing.com will stop neural voices; the offline voices still work. Tick the fallback box to let a conversion continue offline automatically.

“No readable text was found” — the EPUB is image-only (scanned), DRM-protected, or damaged. DRM-protected books cannot be converted.

Some chapters are missing — very short sections (< 350 characters) are merged into the previous chapter, and front-matter/table-of-contents pages are skipped on purpose.

The MP3 shows no chapters — chapter markers are stored as ID3v2.3 CHAP frames; Windows Media Player ignores them, VLC/foobar2000/Poweramp show them.

The MP3 will not play in my car stereo / old player — some devices reject 24 kHz MPEG-2 files. Choose the Compatible (64 kbps) or Maximum (128 kbps) preset: those are standard 44.1 kHz MP3 files and play almost everywhere.

Building from source

Python 3.13 on Windows:

py -3.13 -m venv build\venv
build\venv\Scripts\pip install -r requirements.txt pyinstaller
build\venv\Scripts\pyinstaller --noconfirm --clean --onefile --windowed ^
  --name EpubToMp3 --paths src ^
  --hidden-import win32com.client --hidden-import pythoncom --hidden-import pywintypes ^
  --hidden-import soundfile --hidden-import _soundfile ^
  --workpath build\pyi --specpath build --distpath dist src\launcher.py

Add --icon <path.ico> if you want a custom icon; no icon file ships with the repo.

Source layout:

src/epub2mp3/epub.py      EPUB parsing (stdlib only)
src/epub2mp3/textproc.py  text clean-up, sentence splitting, chunking
src/epub2mp3/engines.py   Edge neural TTS + Windows SAPI, LAME encoding
src/epub2mp3/audio.py     MP3 decoding and polyphase resampling to 44.1 kHz
src/epub2mp3/mp3tool.py   MP3 frame parsing, duration, ID3 chapters, muxing
src/epub2mp3/convert.py   conversion pipeline
src/epub2mp3/gui.py       Tkinter interface
src/epub2mp3/__main__.py  CLI / headless mode

Notes and privacy

• Online voices send the book text to Microsoft's speech service. Offline voices keep everything on your PC — use --offline or pick an offline voice for confidential books.
• This is an independent tool and is not affiliated with Microsoft, Project Gutenberg or any bookstore. Microsoft, Edge and Windows are trademarks of Microsoft Corporation.
• Please respect the copyright of the books you convert.